A number of frequently occurring situations in biochemical and environmental sciences can be viewed as the result of random events involving discrete number of particles in continuous time. Examples range from coupled chemical reactions to predator-prey models. The standard approach to modelling these phenomena is based on simulation, which offers a practical and often intuitive way of grasping the phenomenology of the process. However, stochastic simulations are hard to turn into a statistically rigorous predictive tool, for example for purposes of parameter estimation and posterior inference. In this talk I’ll describe some recent work (joint with Manfred Opper) in which we construct an inferential framework for Markov Jump processes based on a mean-field approximation to the process, in which the couplings between different species are replaced with their averaged effect.